Which Type of Exercise Is More Beneﬁcial for Cognitive Function? A Meta-Analysis of the E ﬀ ects of Open-Skill Exercise versus Closed-Skill Exercise among Children, Adults, and Elderly Populations

: A large number of studies have described a positive relationship between physical exercise and cognition. Physical exercise can be divided into closed-skill exercise (CSE) and open-skill exercise (OSE) based on the predictability of the performance environment. It remains unknown whether either of these types of exercise is more beneﬁcial for cognitive function. Therefore, the purpose of this meta-analysis was to evaluate the e ﬀ ect of OSE versus CSE on cognition. Eligible studies included cross-sectional studies and intervention studies that had a clear deﬁnition of OSE and CSE, and these were used to compare the cognitive performance di ﬀ erences between the two classes of exercise. A total of 15 cross-sectional studies and 4 intervention studies were included in this meta-analysis. Among the cross-sectional studies, the overall e ﬀ ect size for OSE versus CSE was 0.304 (95% conﬁdence interval (CI) ( − 0.097, 1.213); p < 0.05). Further subgroup analysis showed that the overall e ﬀ ect size for OSE versus CSE was 0.247 for inhibition and 0.360 for cognitive ﬂexibility (both p < 0.05). In contrast, no signiﬁcant di ﬀ erences between the two exercise modes were observed in the intervention studies. In particular, there were no signiﬁcant di ﬀ erences in visuospatial attention or in processing speed between the two exercise modes. Taken together, these results suggest that OSE is superior to CSE, especially for executive function, according to the 15 cross-sectional studies examined. However, data from the intervention studies indicate that OSE is not superior. Therefore, additional well-designed, long-term intervention studies are needed to elucidate the potential e ﬃ cacy of OSE in all populations.


Introduction
Physical exercise has the potential to improve several aspects of cognitive function over a lifetime [1], including attention [2], processing speed [3], working memory [4], and executive function [5]. Furthermore, this view is supported by many behavioral and imaging studies [6][7][8][9]. Specifically, physical exercise is conducive to the improvement of children's academic performance [10], the increase of brain-derived neurotrophic factor concentration in adults [11], and the increase of brain volume in the elderly [12]. Cognitive function refers to the process of acquiring or applying knowledge, or the process of information processing [13], which is the most basic psychological process of humans.

Selection Criteria
Studies that met the following inclusion criteria were considered eligible: (1) full-text studies written in English, (2) studies with an intervention or cross-sectional study design, (3) studies with CSE and OSE specifically defined and discussed as independent variables, and (4) studies that identified at least one cognition-related indicator as a dependent variable. After eliminating duplicate or irrelevant articles, the remaining articles were subjected to a full-text evaluation. Further screening was conducted according to the above four criteria, and articles meeting all criteria were included in the meta-analysis. An overview of the selection process and the numbers of included and excluded studies are provided in Figure 1.

Selection Criteria
Studies that met the following inclusion criteria were considered eligible: (1) full-text studies written in English, (2) studies with an intervention or cross-sectional study design, (3) studies with CSE and OSE specifically defined and discussed as independent variables, and (4) studies that identified at least one cognition-related indicator as a dependent variable. After eliminating duplicate or irrelevant articles, the remaining articles were subjected to a full-text evaluation. Further screening was conducted according to the above four criteria, and articles meeting all criteria were included in the meta-analysis. An overview of the selection process and the numbers of included and excluded studies are provided in Figure 1.

Data Extraction and Analysis
The publication year, characteristics of the participants, intervention methods, exercise experience, and cognitive assessment tools were recorded for each cross-sectional and intervention study examined. The number of participants in the OSE and CSE groups, as well as the mean ± standard deviation of the relevant data in the cognitive measurement tasks, were also extracted. For the intervention studies, the mean ± standard deviation data at baseline and post-test were extracted

Data Extraction and Analysis
The publication year, characteristics of the participants, intervention methods, exercise experience, and cognitive assessment tools were recorded for each cross-sectional and intervention study examined. The number of participants in the OSE and CSE groups, as well as the mean ± standard deviation of the relevant data in the cognitive measurement tasks, were also extracted. For the intervention studies, the mean ± standard deviation data at baseline and post-test were extracted from the cognitive measurement tasks. If the standard deviation data were not available, we converted the standard error to a standard deviation [32], or calculated the standard deviation according to the upper and lower limits of the data (e.g., the 95% confidence interval (CI)). Comprehensive Meta-Analysis (CMA) version 3 (Bio-stat Inc., Englewood, NJ, USA) was used to summarize and analyze the data. For multi-term studies involving two or more cognitive tasks, each task was designated a type of cognitive function. Moreover, one task could include multiple subtasks. Over-quantities and inconsistencies of subtasks in evaluating different cognitive functions can lead to biased and unrepresentative results in a meta-analysis. Therefore, in each task, we screened the most representative results as a measure of cognitive function and ensured that each task had the same number of subtasks.
The specific cognitive functions involved in all of the observational studies we screened, including inhibition, cognitive flexibility, visuospatial attention, and processing speed, were classified. Each subfunction was further analyzed using CMA.
For the data analysis, based on the methodology used in other meta-analyses [33,34], we used the standard mean difference as the effect size of overall cognition in each study. The explanation of the effect size estimation is consistent with Cohen's guideline [35], in which 0.2 is a small effect, 0.5 is a medium effect, and 0.8 is a large effect. The significance of the p-value was set at p < 0.05. The statistical homogeneity (I 2 ) of the effect size was assessed using the Cochran's Q test. I 2 is usually used to indicate the proportion of heterogeneity; I 2 is considered low when it is less than 25%, 50%-75% is considered medium, and greater than or equal to 75% is considered high [36].

Risk of Bias Assessment
The generated results were analyzed for publication bias and a corresponding funnel plot was created. The asymmetry or incompleteness of a funnel chart means that publication bias may exist. It can only make a rough qualitative judgment on publication bias; therefore, it needs enough samples to make a judgment [37]. As such, only the 15 cross-sectional studies selected were analyzed for publication bias. The funnel chart produced (with the abscissa representing the effect quantity and the ordinate representing the standard error) showed a small sample bias. When the negative results were removed, the p-value remained significant (p < 0.05).

Study Quality Assessment
Considering the limited number of intervention studies available, only the 15 cross-sectional studies were examined. The quality of each cross-sectional study was rated by two independent reviewers (H.Z. and W.G.) by using a study quality assessment tool developed by Fuzeki et al. [38] and Engeroff et al. [39]. A total of 12 questions were presented, with 12 being the highest score possible. Two reviewers scored each study independently according to the detailed scoring criteria, and differences in scoring were settled by a third reviewer (B.W.).

Comparison of Overall Cognitive Performance in Cross-Sectional Studies
The effects of OSE and CSE on overall cognitive function are summarized in Figure 2 (Raw data for all forest maps can be obtained in Supplementary Materials). The effect size for OSE versus CSE on overall cognition performance was 0.304 (95% CI (−0.097, 1.213); p < 0.05). In addition, no significant heterogeneity was observed across the studies examined (Q(14) = 16.207; p > 0.05; I 2 = 13.62%). In a funnel plot analysis, two of the included studies accounted for the observed asymmetry of the funnel plot(Funnel plots can be obtained from Supplementary Materials). One study [20] was outside the inverted cone and the other [44] was at the edge of the inverted cone. When these two studies were excluded, the significant results originally obtained were not affected, with an effect size of 0.192 (95% CI [−0.097, 0.724]; p < 0.05).

Comparison of Overall Cognitive Performance in Cross-Sectional Studies
The effects of OSE and CSE on overall cognitive function are summarized in Figure 2 (Raw data for all forest maps can be obtained in supplementary materials). The effect size for OSE versus CSE on overall cognition performance was 0.304 (95% CI (-0.097, 1.213); p < 0.05). In addition, no significant heterogeneity was observed across the studies examined (Q(14) = 16.207; p > 0.05; I 2 = 13.62%). In a funnel plot analysis, two of the included studies accounted for the observed asymmetry of the funnel plot(Funnel plots can be obtained from supplementary materials). One study [20] was outside the inverted cone and the other [44] was at the edge of the inverted cone. When these two studies were excluded, the significant results originally obtained were not affected, with an effect size of 0.192 (95% CI [-0.097, 0.724]; p < 0.05).

Specific Cognitive Performance
In each study, specific cognitive performance was selected as the evaluation index. Considering the small number of intervention studies selected, a subgroup analysis was conducted only for the 15 cross-sectional studies. Studies measuring the same type of cognitive ability were used to explore specific subfunctions of CSE and OSE to fully understand the ability of these two modes of exercise to promote cognitive performance. It is worth noting that each study adopted the same or similar paradigm for the measurement of specific subfunctions. The final results are presented in Table 3 Raw data for tables 3, 4 can be obtained in the supplemental materials. The overall effect size for OSE versus CSE on inhibition was 0.247 (95% CI (-0.173, 1.213); p = 0.042). In addition, no significant heterogeneity was observed among the studies (Q(6) = 10.529; p > 0.05; I 2 = 43.01%). Regarding cognitive flexibility, the combined effect size was 0.360 (95% CI (0.036, 0.923); p = 0.013), and no significant heterogeneity was observed (Q(4) = 5.382; p > 0.05; I 2 = 25.68%). Furthermore, no significant differences in visuospatial attention or processing speed were found between OSE and CSE.

Specific Cognitive Performance
In each study, specific cognitive performance was selected as the evaluation index. Considering the small number of intervention studies selected, a subgroup analysis was conducted only for the 15 cross-sectional studies. Studies measuring the same type of cognitive ability were used to explore specific subfunctions of CSE and OSE to fully understand the ability of these two modes of exercise to promote cognitive performance. It is worth noting that each study adopted the same or similar paradigm for the measurement of specific subfunctions. The final results are presented in Table 3 Raw  data for Tables 3 and 4 can be obtained in the Supplementary Materials. The overall effect size for OSE versus CSE on inhibition was 0.247 (95% CI (−0.173, 1.213); p = 0.042). In addition, no significant heterogeneity was observed among the studies (Q(6) = 10.529; p > 0.05; I 2 = 43.01%). Regarding cognitive flexibility, the combined effect size was 0.360 (95% CI (0.036, 0.923); p = 0.013), and no significant heterogeneity was observed (Q(4) = 5.382; p > 0.05; I 2 = 25.68%). Furthermore, no significant differences in visuospatial attention or processing speed were found between OSE and CSE.

Comparison of Overall Cognitive Performance in Intervention Studies
A total of four intervention studies were analyzed, one of which adopted a within-subjects design [21]. Due to the small number of studies that met our criteria, we did not exclude the latter study. According to the forest plot shown in Figure 3, no significant difference in overall cognitive performance was observed between CSE and OSE.

Comparison of Overall Cognitive Performance in Intervention Studies
A total of four intervention studies were analyzed, one of which adopted a within-subjects design [21]. Due to the small number of studies that met our criteria, we did not exclude the latter study. According to the forest plot shown in Figure 3, no significant difference in overall cognitive performance was observed between CSE and OSE.

Moderator Analysis
The results of a moderator analysis of OSE versus CSE are summarized in Table 4. To analyze the influence of potential moderator variables on overall cognitive function, and considering that only a small number of intervention studies were available, we only analyzed the moderator variables of the 15 cross-sectional studies according to age group. These groups included children aged 5-16 years, young adults aged 17-35 years, and elderly aged >56 years. Our goal was to explore whether OSE and CSE influence the overall cognitive function of different age groups. Based on our selection criteria, we had one study of children, eight studies of young adults, and six studies of the elderly to evaluate. Consequently, we only compared differences for the younger and older adults. The effect size for OSE versus CSE on overall cognition in the young adults was 0.384 (95% CI (-0.097, 1.213); p = 0.002) ( Table 4), and no significant heterogeneity was observed across these studies (Q(7) = 8.129, p > 0.05, I 2 = 13.89%). Meanwhile, the effect size for OSE versus CSE on overall cognition in the older adults was 0.197 (95% CI (0.033, 0.923); p = 0.105) ( Table 4). There was no significant heterogeneity observed between these studies either (Q(5) = 6.077, p > 0.05, I 2 = 17.72%). Taken together, these results demonstrate that the cognitive benefits of OSE are greater for younger adults.
According to a previously described study quality assessment tool [38,39,53], each study was graded. Scores for the 15 cross-sectional studies ranged from 7 to 9. A score of 9 was considered to indicate high quality, 8 indicated moderate quality, and 7 indicated low quality(Detailed grading rules and scores for each article can be obtained in the supplemental materials). Differences in the effects of OSE and CSE on cognitive function were found to be significant in both the high-and moderate-quality studies and were not significant in the low-quality articles.

Moderator Analysis
The results of a moderator analysis of OSE versus CSE are summarized in Table 4. To analyze the influence of potential moderator variables on overall cognitive function, and considering that only a small number of intervention studies were available, we only analyzed the moderator variables of the 15 cross-sectional studies according to age group. These groups included children aged 5-16 years, young adults aged 17-35 years, and elderly aged >56 years. Our goal was to explore whether OSE and CSE influence the overall cognitive function of different age groups. Based on our selection criteria, we had one study of children, eight studies of young adults, and six studies of the elderly to evaluate. Consequently, we only compared differences for the younger and older adults. The effect size for OSE versus CSE on overall cognition in the young adults was 0.384 (95% CI (−0.097, 1.213); p = 0.002) ( Table 4), and no significant heterogeneity was observed across these studies (Q(7) = 8.129, p > 0.05, I 2 = 13.89%). Meanwhile, the effect size for OSE versus CSE on overall cognition in the older adults was 0.197 (95% CI (0.033, 0.923); p = 0.105) ( Table 4). There was no significant heterogeneity observed between these studies either (Q(5) = 6.077, p > 0.05, I 2 = 17.72%). Taken together, these results demonstrate that the cognitive benefits of OSE are greater for younger adults.
According to a previously described study quality assessment tool [38,39,53], each study was graded. Scores for the 15 cross-sectional studies ranged from 7 to 9. A score of 9 was considered to indicate high quality, 8 indicated moderate quality, and 7 indicated low quality(Detailed grading rules and scores for each article can be obtained in the Supplementary Materials). Differences in the effects of OSE and CSE on cognitive function were found to be significant in both the high-and moderate-quality studies and were not significant in the low-quality articles.

Discussion
To the best of our knowledge, this is the first meta-analysis of all age groups that compared the effects of CSE and OSE on cognitive performance based on results obtained from cross-sectional and intervention studies. Among the 15 cross-sectional studies examined, OSE was superior to CSE with a small effect regarding cognitive performance. Additional subgroup analyses further demonstrated that OSE led to positive effects on inhibition and cognitive flexibility compared to CSE. Meanwhile, visuospatial attention and processing speed did not exhibit significant differences between the two modes of exercise. Since inhibition and cognitive flexibility are both important subfunctions of executive function, these results provide preliminary support that OSE can achieve better executive function than CSE.
In contrast, among the four intervention studies examined, no significant differences were observed between OSE and CSE. Moreover, while the direction of the effect size was biased toward OSE, the associated p-value was not significant. Therefore, our research results only partially support the hypothesis that OSE is superior to CSE in terms of executive function.

Differences in Cognitive Function between OSE and CSE
Consistent with the findings of a recent systematic review [53], we found OSE to be more effective at improving cognitive performance than CSE. Furthermore, when we refined cognitive performance, OSE produced better inhibition and cognitive flexibility than CSE. In two previous studies [43,50], event-related potential was used to compare OSE and CSE, and the former exhibited a better electrophysiological performance. These results are also consistent with the present findings.
Regarding visuospatial attention, previous studies have produced conflicting views. For example, when Giglia et al. focused on the lateralization of athletes' visuospatial attention, no significant difference in visuospatial attention was observed between the CSE and control groups [40]. Moreover, the performance of both groups was worse than that for OSE, suggesting that a stable sports environment may not effectively train individuals to distract their visuospatial attention. In contrast, another study conducted by Chueh et al. showed no difference in visuospatial attention between OSE and CSE [47]. We are more inclined to support the former result based on our understanding of the two exercise modes. It should be noted that data for the present meta-analysis exhibited a relatively large effect size that was biased toward OSE. However, the results were not significant. There are two possible reasons for the latter observation. One, with only three studies included in our analysis, our results are based on a very small sample size. Second, the subjects evaluated for visuospatial attention were mostly young adults. Thus, the absence of children and the elderly as study subjects may also have contributed to our insignificant results. The results of the present study also confirmed that no significant difference exists between the two exercise modes regarding the performance of processing-speed-related tasks [48].
To further enrich and support the findings of the cross-sectional studies examined, we analyzed four intervention studies. The main reason for incorporating the intervention studies was to facilitate causal inference. Of the four articles we examined, two [24,52] provided support that OSE is better for cognitive promotion, one study showed that OSE and CSE have unique advantages for specific cognitive subfunctions [23], and the final study showed no difference between the two exercise modes at the behavioral level [21]. The results of the present meta-analysis indicate that there was no significant difference in the promotion of cognitive ability between OSE and CSE. Possible reasons for this result are that there were not many intervention studies available that have investigated the effects of these two modes of exercise on cognition, and only one of the four papers we examined could cause a large error in the meta-analysis. Therefore, in future studies, it will be important to include a greater number of intervention studies to ensure that reliable results are obtained. Second, the cognitive improvement from exercise appears to be related to intervention duration. This finding is supported by a recent systematic review in which it is demonstrated that only long-term exercise plays a positive role in brain function and structure [18]. For our meta-analysis, we included three long-term exercise studies [23,24,52], as well as a short-term exercise study [21]. Thus, a potential cause of our inconclusive results may be that we included all four studies for analysis. However, due to the limited number of articles included, it is not feasible to judge the correctness of this conjecture based on the moderate analysis grouped by intervention duration. Thus, additional studies are needed. It is worth mentioning that a recently published study shows that CSE provides more obvious advantages for retrospective memory than OSE after acute exercise [54]. A possible reason for the inconsistency may be mainly due to the different experimental paradigms. In the future, more intervention studies are needed to further reveal the different effects between the two exercise modes.
In summary, the results of this meta-analysis show that OSE was more effective than CSE in promoting the development of cognitive function. Moreover, this difference was reflected in higher cognitive processes, namely executive function.

Potential Mechanism for OSE's Superiority to CSE
In the present meta-analysis, OSE was found to have a greater benefit for cognitive function than CSE. However, the effect differed according to specific cognitive domains. In particular, OSE was more likely to promote inhibition and cognitive flexibility. We hypothesize that this result was due to participants' need to mobilize more cognitive resources in self-control and transformation to adapt to changes in the external information and the multisensory environments of OSE [55]. In contrast, CSE involves a relatively predictable and stable environment. Thus, fewer cognitive resources need to be mobilized during this mode of exercise. Inhibition and cognitive flexibility are important components of executive function. In OSE, this high-level cognitive process could be improved. In terms of processing speed, since the measurements of processing speed are based on relatively simple tasks, the cognitive resources required for mobilization are very limited. Therefore, no significant differences between OSE and CSE were observed when considering tasks requiring a low cognitive level [48,56].
Regarding visuospatial attention, our meta-analysis did not identify any differences between the two exercise modes. Exercise in a complex environment has been shown to increase the cortical thickness, neurogenesis, and heighten neurotransmission [57,58]. Moreover, a prior study found that prefrontal cortex activation is much greater in children than in adults [59]. In addition, the prefrontal cortex plays an important role in resisting interference and maintaining targets. Thus, the effects of different exercise modes on visuospatial attention vary in different age groups. To date, very few studies have included children and the elderly in the evaluations of modes of exercise. Consequently, a lack of comprehensive coverage of the characteristics of these subjects may account for the insignificant results we obtained. Moreover, the latter partly explains current inconsistencies among views of the two exercise modes regarding visuospatial attention.
From the perspective of educational applications, the current evidence indicates that OSE was more beneficial to the cognitive development of normal people, and these findings will be helpful to the replanning and arrangement of physical education for education decision-makers. In terms of clinical application, physical exercise has also been proved to be an effective means to improve mild cognitive impairment and dementia [60]. Therefore, we speculate that the unique advantage of OSE in cognitive ability may also exist in special groups.
Our results suggest that, compared with CSE, OSE had more advantages in improving cognitive function at the behavioral level. It provides some implications for people's choice of exercise modes. This existing finding could be further extended to the level of brain mechanisms in the future.

Strengths and Limitations
A major advantage of our meta-analysis is that it provides an up-to-date summary of differences in the effects of CSE and OSE on cognitive performance without limitations regarding the experimental design methods, age of subjects, and publication date. However, the conclusions of this meta-analysis should be considered regarding the following limitations. First, only 4 out of the 19 studies examined were intervention studies. Thus, our meta-analysis of the intervention studies was not sufficiently reliable, and the conclusions can only provide some references and suggestions for future research. Moreover, future studies need to cautiously consider causal inference. Second, the cross-sectional studies examined did not provide specifics regarding exercise time, frequency, and specific task performance for each of the participating genders. Therefore, the number of regulatory variables was not sufficient for further analysis, and this was not conducive for explorations of sources of heterogeneity. Third, while our moderator analysis indicated that OSE provided greater cognitive benefits to young adults, there have only been a few studies that investigated children participating in OSE. Therefore, it remains for future studies to confirm whether OSE is beneficial for children.
Finally, we only used behavioral data to determine differences in cognitive performance between the two exercise modes. In future studies, it will be important to employ functional magnetic resonance imaging (fMRI) technology to explore functional and structural changes in the brain that affect cognition as a result of the two exercise modes. Key considerations will be which exercise mode is more helpful in promoting and updating current inference and which neural mechanism(s) underlie the changes observed. Additionally, our search strategy was limited to full-text studies published in English. Future studies should include all relevant literature, independent of language, to have a complete dataset that is not biased according to language.

Conclusions
Compared with CSE, OSE improved cognitive function more effectively, regardless of baseline age. The benefits of OSE were especially apparent in inhibition and cognitive flexibility, and these should potentially promote better executive function. In addition, the results of the moderator analysis showed that the cognitive benefits of OSE were greater in young people. However, it is important to interpret the present results with caution due to the limited number of studies available. It is recommended that long-term intervention studies, as well as the use of fMRI, should be used to uncover the underlying brain mechanisms by which OSE affects cognitive function.

Conflicts of Interest:
The authors declare no conflict of interest.